The information your metadata fields need is usually already on the asset: in the file name, the folder it sits in, the metadata uploaded with the original file, the transcript or the comments. AI Metadata Enrichment gets it from there into your fields. For each field, you write a short prompt that says what you're looking for. Users then pick an asset, or several, choose the fields to fill, and Iconik suggests a value for each one. If the answer isn't on the asset, the field stays empty.
Users can review each suggestion before it saves, accept a whole batch at once, or have suggestions saved straight away with auto-accept. Every value saved this way is marked as AI-generated, so you can always tell it apart from values entered by people. As an admin, you decide who can use it and how well it understands your fields. This page is for admins. For generating and reviewing suggestions day to day, see AI Metadata Enrichment User guide.
Roles
- You need to have Metadata Administration roles configured
- can_read_jobs
- can_write_jobs
Set up in five steps
Step 1: Check your AI settings
AI Metadata Enrichment is on by default for accounts that have accepted the AI terms. Only a Domain Owner can change AI settings. Your inputs and outputs are not used to train AI models, and processing stays in your account's region.
- Ask your Domain Owner to go to Admin > Settings > System Settings > AI.
- Check that Allow AI features and AI Metadata Enrichment are both on.
If AI was switched off earlier, the AI terms need to be accepted again first.
Step 2: Check who already has access
The Domain Owner has access straight away, so the feature can be set up and tested before it's shared with the team in Step 5. Whether other users already have access depends on if they're in a role groups:
| User | Access |
|---|---|
| Standard or Power user in at least one role group | No access until you give their role group the roles in Step 5 |
| Standard or Power user who isn't in any role group | Access straight away, since they have every capability of their user type |
| Browse-Only user | No access. Role groups can only limit what a user type allows, not add to it. |
Note: If some of your users aren't in a role group, they can try the feature before you've set up your fields. If you'd rather they waited, add them to a role group in Admin > Role Groups.
Step 3: Tell the AI what each field is for
This is the step that matters most for the quality of the suggestions.
To set up a field:
- Go to Admin > Metadata > Fields (tab) and open a field.
- Add a Description, used as a tooltip in the UI.
- Add an AI Prompt, used by the AI.
- Click Update.
The AI reads three settings on each metadata field:
| Setting | Who sees it | What to write |
|---|---|---|
| Label | Everyone, as the field name | Use a clear descriptive name. |
| Description | Users see it when they hover over the field | A short, plain explanation of what the field holds. The AI uses it as backup if there's no AI Prompt. |
| AI Prompt | Only the AI, not displayed in the UI to end users | Specific instructions: format, length, language, where to look, and what to do when there isn't enough information. Takes priority over the Description. |
If a field has no Description and no AI Prompt, users will see a warning icon (⚠) next to it before they generate, so they know the suggestion is unlikely to be useful.
What makes a good AI Prompt
- Say what you want, precisely. "A 50-word summary in English" works better than "Summary".
- Point to the source when you know where the answer is: the transcript, the file name, the folder path.
- Set the format. Lowercase, comma-separated, date format, number of items.
- Say when to leave it blank. Asking the AI to leave a field empty rather than guess keeps your metadata clean.
- Ask only for what's in the context. The AI works from the information on the asset. It doesn't analyse what's visible in an image or video, unless that's already described in a transcript, tags or other metadata.
Example prompts
The AI looks at all the information available on the asset by default. If you know where the answer is, you can tell the AI where to look. Each example below points the AI to the sources to use. Copy one into a field's AI Prompt and adjust it to your own naming and content.
| Where the answer is | Example AI Prompt |
|---|---|
| File name |
Extract the project ID from the filename. Rules: |
| Folder or collection path |
Extract the brand from the collection path. Rules: |
| Transcript (description) |
Write a one-sentence description of this video based on the transcript. Rules: |
| Transcript (keywords) |
Generate up to 8 precise keywords for this asset. Include: Avoid: |
| Drop-down field |
Choose the best content type from this list: Rules: |
Field types
- Drop-down fields: the AI picks from your existing options.
- Number fields with a minimum or maximum: suggestions outside the range aren't saved. The other fields in the job are still saved.
- Date fields: say which date you mean (for example, recording date rather than upload date) and where to find it.
Build up metadata in passes
You can run enrichment on the same asset more than once. Every value it fills becomes something the next run can draw on. For example, do an early pass from the file name, the metadata uploaded with the original file and the collection path, then run it again later for other fields as the asset moves through your workflow.
Which fields to start with
Start with fields that people fill in by hand often, that follow clear rules, and that can be worked out from a transcript, the file name, the folder path or existing metadata. Fields that rely on human judgement, such as an editorial rating, are better left to people.
Step 4: Test on your own content
Before you give access to your team, try the feature on a few assets you know well. The steps below are the short version. For the full walkthrough of generating and reviewing suggestions, see AI Metadata Enrichment User guide.
- Pick an asset with a transcript and some existing metadata.
- Open the asset and click Generate metadata with AI in the Metadata view, or in the ⋯ menu.
- Check for fields with a warning icon (⚠). These have no Description or AI Prompt yet.
- Choose the fields to fill and start the job. Leave Auto-accept off for testing, so you can see every suggestion.
- When the job is done, open the review. Current values are on the left, suggestions on the right.
- If a suggestion isn't what you expected, discard it, adjust the field's AI Prompt, and generate again.
Once single assets look good, try a small bulk run, for example 10 to 20 assets from a collection, before running it on larger sets.
Step 5: Roll out to your team
When you're happy with the results, give a small group access first and widen it as you gain confidence.
Give a role group access
- Go to Admin > Role Groups and open the role group.
- Switch to Advanced view.
- Turn on each role in Roles a user needs below. The metadata roles are under Metadata Values, the job roles under Jobs.
- Click Update.
Roles a user needs
To generate and save suggestions, a user needs all of these roles. Standard and Power users who aren't in a role group already have them.
| Role | What it allows | Available for |
|---|---|---|
| Can trigger metadata ai filling | Starting AI Metadata Enrichment on assets | Standard and Power users |
| Can read metadata values | Seeing metadata values | All user types |
| Can write metadata values | Editing metadata, which is needed to save suggestions | Standard and Power users |
| Can read jobs | Seeing jobs, so the user can follow their AI jobs | All user types |
| Can write jobs | Updating jobs | Standard and Power users |
A user in several role groups gets the roles of all of them combined. Leave Can see all jobs off, since it shows every job in the account. See Roles for more.
Access to the content
Users also need write access to the metadata view they want to fill, and to the assets:
- If they can only read a view, the job fails with "User doesn't have write access on the view".
- In a bulk selection, assets they can't edit are skipped.
Then
- Share AI Metadata Enrichment User guide with them.
- Agree on how you'll use auto-accept (see below).
- Ask users to tell you about suggestions that were wrong or missing, and about fields showing a warning icon. Their feedback is how you improve the AI Prompts over time.
Reviewing versus auto-accepting
Users can save suggestions in three ways:
| Method | What happens | Best for |
|---|---|---|
| Review per asset | The user opens each asset, then accepts, edits or discards each suggestion | Summaries, keywords and other free text, where domain knowledge matters |
| Accept all pending | In a bulk job, the user saves all remaining suggestions at once | Large batches after checking a sample set |
| Auto-accept | Turned on before the job starts. Suggestions are saved without review | Empty fields that come from technical information (file name, folder path) or fixed drop-down options |
Important: Auto-accept replaces existing values. If a user turns on auto-accept and includes fields that already have a value, the AI suggestion replaces that value, without anyone reviewing it. Ask your team to turn on Show empty fields only when using auto-accept, so only empty fields are filled. Anyone with access can use auto-accept.
Track results and troubleshoot jobs
AI Metadata Enrichment runs as jobs. Each time someone generates suggestions, Iconik creates a job. A bulk run is one parent job with a sub-job for each asset.
Go to Admin > AI Metadata Enrichment to follow the progress and results of AI jobs from every user in your account. You can open anyone's job, but only the person who started it can accept, discard or regenerate its suggestions.
If something goes wrong, go to Admin > Jobs to troubleshoot. Open the job to see its status and, if it failed or was skipped, the reason. For more information go to Jobs Overview.
How AI-filled values are marked
Every saved value records where it came from, so you can tell AI values from values entered by people:
| Marked as | Meaning |
|---|---|
| AI | Saved with auto-accept, not reviewed by anyone |
| AI accepted | Reviewed and accepted by a user; including “Accept all pending” |
| (no mark) | Entered or last changed by a person |
If someone edits an AI value later, it counts as entered by a person.
Troubleshooting
Users can't see Generate metadata with AI.
Check that AI features are allowed (Step 1) and that the user's role groups have all the roles under Roles a user needs (Step 5).
A field came back empty.
The AI didn't find enough information to suggest a value. This is expected. Add a transcript, or adjust the AI Prompt and try again. Sometimes too strict prompts will hinder the feature to produce and answer.
The job was skipped.
All the selected fields already had values, or there wasn't enough information. Open the job to see the reason.
Some assets in a bulk job weren't processed.
Assets the user can't edit are skipped. Check their access to those assets.
FAQ
Which plans include AI Metadata Enrichment?
All plans. It uses AI credits.
How many AI credits does AI Metadata Enrichment use?
AI Metadata Enrichment uses the same AI credits as Iconik's other AI features, shared across your whole account. See AI Credits for what's included in your plan and how usage is calculated.
Are credits used again if the same asset is run more than once?
Yes. Credits are counted for every run of every asset. If the same asset is run three times, it counts as three assets.
Can I limit how many credits users spend?
There's no spending cap per user. To keep costs predictable, test on small batches and agree with your team before anyone runs large collections or saved searches. You can also choose who gets access (Step 5).
What information does the AI use?
Everything Iconik already knows about the asset:
- Always: the asset title and type, the created and modified dates, the full path of the collections it's in, and user info.
- When available: existing metadata, the original file name, technical metadata (such as format, size, resolution and codec), the transcript, subtitles, time-based segments (such as scenes, shots and people identified by face recognition), comments, and related assets.
Does it analyse what's visible in images or video?
No. The AI works from the information on the asset, so visual details are only used if they're already described in a transcript, tags or other metadata.
Can one field use the AI suggestion for another field in the same run?
Yes. Name the other field in the AI Prompt, and select both fields when generating. For example:
- Summary: "An English summary of the transcript"
- FR Summary translation: "A French translation of the Summary field"
The Summary is filled first, and its suggestion is then reused for the translation. If the prompt doesn't name the other field, for example "A summary in French of the transcript", the AI writes a new summary from the transcript instead of translating the Summary field.
Is there a limit on how many assets I can run at once?
No. Users can run it on a selection of assets, a whole collection or a saved search.
What happens to suggestions nobody reviews?
They're removed after 30 days. The AI credits for those runs are still used, so ask your team to review or discard their jobs.
Can I limit auto-accept to certain users, or choose which metadata views get suggestions?
Not at the moment. Anyone with access can use auto-accept, on any metadata view they can edit. To control this, choose carefully who gets access (Step 5).
Setup checklist
- [ ] AI features allowed and AI terms accepted by a Domain Owner
- [ ] AI Metadata Enrichment on in Admin > Settings > System Settings > AI
- [ ] Description and AI Prompt added to your fields
- [ ] Tested on a few single assets and a small bulk run
- [ ] All roles under Roles a user needs given to your first role group
- [ ] Agreed with your team how to use auto-accept
- [ ] Team pointed to the user guide and asked for feedback